Research On Inter-frame Compression Algorithm For 3D Point Cloud

With the great progress of 3D sensing and 3D reconstruction technology,point cloud,as the mainstream 3D representation format,has become one of the new generation of digital media after audio and video.Among them,single-frame static point cloud are widely used in digital cultural relics protection,geographic information systems and other fields,and dynamic point cloud sequences play an important role in scenarios such as automatic driving and unmanned inspection.Point cloud often contain massive data points,and each point contains various types of attributes.The huge amount of data brings severe challenges to the storage and transmission of multimedia systems.Therefore,the research on the point cloud information coding algorithm is of great significance.However,for dynamically acquired point cloud sequences,existing research mainly uses spatial correlation for de-redundancy,and the utilization of temporal correlation is not sufficient,and there is still a large room for performance improvement in point cloud inter-frame coding.In this thesis,a geometry interframe coding algorithm based on node sparsity is proposed for the inter-frame coding of geometry information of point cloud;an attribute inter-frame coding algorithm based on spatial distance is proposed for the inter-frame coding of attribute information of point cloud.The proposed algorithms effectively improve the compression performance of point cloud from the perspective of geometry coding and attribute coding respectively.For the inter-frame coding of geometry information based on multi-branch tree in the AVSPCC coding framework,the single point inter-frame coding algorithm and the occupacy information inter-frame coding algorithm are studied respectively in this thesis.Firstly,by analyzing the problem of the low accuracy of the single point direct coding algorithm in the existing framework,an algorithm to determine the coding qualification of single point by using the information of the corresponding nodes in the reference frame is proposed.Then,the problem that the two sets context models in existing framework having different performances for point cloud of different densities is analyzed,and an adaptive selection algorithm for inter-frame context models based on the density of point cloud slice is proposed.Finally,the correlation between the number of points in the corresponding node in the reference frame and the occupancy pattern of nodes to be encoded is analyzed,and a occupancy information coding algorithm based on the number of points in the node is proposed.The experimental results show that,compared with the existing inter-frame coding branch of AVS-PCC,under the lossy coding condition,the eligibility judgment algorithm of single point of inter-frame brings a performance improvement of 0.8% on average,the adaptive selection of inter-frame context model algorithm brings a performance gain of 1.1%on average,and the occupancy information coding algorithm based on the number of points in the node brings a performance gain of 0.4% on average.For the point cloud attribute information coding in the AVS-PCC coding framework,this thesis firstly extends the attribute intra prediction coding algorithm in the coding framework,and uses the temporal correlation of the point cloud to de-redundant attribute information.On this basis,this thesis further conducts a statistical analysis on the prediction residuals of intra-frame and inter-frame modes,an attribute predictive coding algorithm for adaptive selection of intra-frame and inter-frame prediction modes is proposed.Finally,for the coding problem of duplicate points in the reference frame and the current frame,a weighted predictive coding algorithm is proposed,which comprehensively considers the attribute values of inter-frame duplicate points and non-duplicate points.The experimental results show that,compared with the existing attribute intra coding scheme in AVS-PCC,under the condition of lossy coding,the adaptive attribute inter-frame predictive coding algorithm brings a performance gain of 15.2% on average,the attribute weighted predictive coding algorithm for inter-frame duplicate points brings a performance gain of 17.7% on average.